See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/283210919

A statistical analysis of EV charging behavior in

the UK
Conference Paper October 2015
DOI: 10.1109/ISGT-LA.2015.7381196

CITATIONS

READS

200

3 authors, including:
Jairo Quiros-Tortos

Luis(Nando) Ochoa

The University of Manchester

The University of Manchester

32 PUBLICATIONS 55 CITATIONS

174 PUBLICATIONS 1,719 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

SEE PROFILE

Available from: Luis(Nando) Ochoa

Retrieved on: 10 May 2016

Accepted Paper

A Statistical Analysis of
EV Charging Behavior in the UK
Jairo Quirs-Torts
The University of Manchester
Manchester, UK
jairoquirostortos@ieee.org

Luis F. Ochoa
The University of Manchester
Manchester, UK
luis_ochoa@ieee.org

AbstractTo truly quantify the impact of electric vehicles (EVs)

on the electricity network and their potential interactions in the
context of Smart Grids, it is crucial to understand their charging
behavior. However, as EVs are yet to be widely adopted, these
data are scarce. This work presents results of a thorough statistical analysis of the charging behavior of 221 real residential EV
users (Nissan LEAF, i.e., 24kWh, 3.6 kW) spread across the UK
and monitored over one year (68,000+ samples). Probability
distribution functions (PDFs) of different charging features (e.g.,
start charging time) are produced for both weekdays and weekends. Crucially, these unique PDFs can be used to create stochastic, realistic and detailed EV profiles to carry out impact
and/or Smart Grid-related studies. Finally, the effects of the EV
demand on future UK distribution networks are discussed.
Index Terms-- Electric vehicles, real data, statistical analysis.

I.

INTRODUCTION

The increasing adoption of electric vehicles (EVs) in the

UK is expected to increase given their potential contribution to
reduce greenhouse gases and dependency on fossil fuels [1].
This, nonetheless, may significantly stress the electricity network. If integrated efficiently, however, EVs could be used as
a crucial resource in the context of Smart Grids [2]. To truly
quantify the corresponding impacts or benefits it is, nonetheless, critical to understand the charging behavior of EV users.
However, as EVs are yet to be widely adopted, these data are
currently scarce or limited to small samples [3].
The demand from EVs is a stochastic variable difficult to
model given that it depends on customer behavior. Initial
works have used Travel Survey data (e.g., [4]) to understand
how consumers drive traditional combustion engine vehicles
to estimate how they might drive an EV. However, EV users
may exhibit new behaviors given that they are capable of refilling (charging) the battery at home, instead of going to petrol stations. Recent studies have used data from small-scale
EV trials to model the EV demand (e.g., [3, 5, 6]). These studies have achieved their goal; however, the EV models produced may be limited to the particular set of EV users which
were used to create them.
Therefore, it is clear that EV data from large-scale trials is
needed to truly understand the charging behavior of a more
diverse population of EV users. This in turn will benefit the
corresponding analysis required to truly quantify the impact of

Becky Lees
EA Technology Limited
Chester, UK
becky.lees@eatechnology.com

EVs on the electricity network (e.g., [7]) and their potential

interactions in the context of Smart Grids (e.g., [6]).
This work provides a set of results from a thorough statistical analysis of more than 200 Nissan LEAFs [8] used by
residential UK customers (i.e., 24 kWh battery capacity, 3.6
kW demand), which have been monitored during the My
Electric Avenue project [9]. More than 68,000 charging
events have been recorded over one year since Dec 2013. For
each EV charging event, the onboard monitoring system records the start time, end time, initial state of charge (SOC), and
final SOC [8]. Probability distribution functions (PDFs) are
presented here from these metrics. No significant variance in
the charging behavior of EVs across seasons was found;
charging across all four seasons is considered. The analysis
includes both weekdays and weekends. By combining the data
from these unique PDFs, researchers can create stochastic,
realistic and detailed profiles to adequately model the EV demand in a straightforward manner. Finally, the potential implications of EV demand on UK networks are discussed.
This paper is organized as follows. Section II creates the
PDFs for the charging metrics required to model the EV demand. Section III presents the methodology to create stochastic, realistic and detailed profiles to adequately model the EV
demand. Section IV analyzes the implications the EV adoption
may have in the aggregated demand in UK networks, particularly in residential low voltage networks. A discussion is provided in section V and conclusions are drawn in section VI.
II.

PROBABILITY DISTRIBUTION FUNCTIONS

This section details the creation of weekday and weekend

PDFs for the number of connections per day, the start charging time, the initial SOC, and the final SOC.
It is known that EV users may require a period of time to
familiarize themselves with the EV and establish their own
charging needs. Indeed, an initial analysis found that charging
behavior shows a more predictable pattern after one week, i.e.,
EV users understand how the battery level meets their driving
requirements. Hence, the corresponding charging events
(< 2%) are excluded from the analysis below.
A. Number of Connections per Day
The number of connections per day presented in Table I
highlights that circa 70% of the EVs are connected only once

This work has been funded by EA Technology Limited, UK, through the
Ofgems Low Carbon Networks Fund Tier 2 Project My Electric Avenue,
2013-2015.

Accepted Paper

Accepted Paper
TABLE I.
No. Conns
Weekday
Weekend

PDF OF THE NUMBER OF CONNECTIONS PER DAY (%)

1
71.26
68.99

2
21.15
21.51

3
5.41
6.62

4
1.51
1.90

5
0.44
0.63

6
0.14
0.24

7+
0.09
0.11

First Connection

Probability (%)

2
1
0
3
Second Connection
2
1
2h

4h

6h

8h 10h 12h 14h 16h 18h

Time of Day - 15 min resolution

20h 22h

24h

Figure 1. PDF of the start charging time per connection Weekday.

2
First Connection

Probability (%)

0
4
Second Connection

3
2
1
0
0h

2h

4h

6h

Units
0
1
2
3
4
5
6
7
8
9
10
11
12

0
0h

TABLE II.

8h 10h 12h 14h 16h 18h

Time of Day - 15 min resolution

20h 22h

24h

Figure 2. PDF of the start charging time per connection Weekend.

a day, irrespective of weekday or weekend. Previous EV studies have not explored multiple charging events, and as such
this finding is unique. Although this may not have a significant impact in the magnitude of the evening peak (as detailed
in section III), it does affect the morning peak as well as the
overall energy consumption. Since three or more connections
are unlikely, and for simplicity, only two connections are studied below (the second is the aggregation of the rest).
B. Start Charging Time
Fig. 1 and Fig. 2 show the PDF per connection of the start
charging time for weekday and weekend, respectively. As
expected, EV users vary their start charging time. Overall, the
first EV connection may occur any time during the day. Nonetheless, a second is more likely to occur after midday. During
weekdays, the first connection usually starts around 8h (before
work) or 18h (after work); the second connection typically
starts at 18h. This highlights that a number of EVs are charged
at home before and after work. During weekends, the first
connection usually starts between 9h and 18h and the second
later in the evening. No significant differences were found
among weekdays (i.e., MondayFriday) and among weekends,
though this is not shown in Fig. 1 and Fig. 2.
C. Initial and Final SOC
The initial and final SOC depend on the number of connections and the time of the day. For instance, an EV charged

PDF OF THE INITIAL AND FINAL SOC PER CONNECTION (%)

Weekday
Initial SOC
Final SOC
1st
2nd
1st
2nd
0.57
0.91
0.00
0.05
3.52
4.23
0.15
0.28
8.38
7.55
0.39
0.96
11.75
9.59
0.74
0.89
11.86
9.51
0.78
1.21
10.87
9.34
1.27
1.42
11.62 11.17
2.07
2.48
12.21 10.51
2.58
3.18
9.46
8.54
3.55
3.63
6.56
6.79
7.05
6.31
6.08
7.91
7.34
9.48
4.03
6.94
5.16
6.46
3.09
7.01
68.92 63.65

Weekend
Initial SOC
Final SOC
1st
2nd
1st
2nd
0.67
1.21
0.03
0.02
3.74
5.28
0.10
0.44
7.45
8.68
0.53
0.81
9.45
9.91
0.62
0.79
10.09 9.66
0.84
1.01
10.31 8.93
1.47
1.63
10.99 9.54
1.93
2.56
11.80 9.76
2.65
2.81
9.59
9.59
3.80
3.80
7.63
7.22
7.28
6.53
7.83
7.52
8.75
12.50
5.99
6.51
6.20
8.53
4.46
6.19 65.80 58.57

overnight that is used in the morning for a short trip will have
a relatively high initial SOC for the next charge. However,
initial analysis highlights that time-dependency is not significant; then, this work focuses on the number of connections.
Table II shows the PDF per connection of the initial and
final SOC during weekdays and weekends. The Nissan LEAF
(24kWh) represents the SOC in 12 units/segments (2kWh per
segment, i.e., 1 segment equals 8.3% of battery capacity). Irrespective of weekday or weekend, Table II highlights that the
first connection starts in more than 70% of the EVs when their
initial SOC is between 3 and 9 segments (i.e., 25 to 75%). It
can also be seen that second connections occur with higher
SOC. In terms of final SOC, Table II highlights that approximately 65% of the EVs finish their first connection with full
battery. Table II finally shows that during weekends, disconnections are more frequent before EVs are fully charged.
III.

CREATION OF EV PROFILES

A. EV Demand and Power Factor

To create daily time-series EV profiles (see section III-B),
it is important to understand the typical EV demand (in kW)
as well as its power factor. To determine these features, the
active, reactive, and apparent power monitored on a specific
EV over a period of four months are used. Fig. 3 shows the
monitored apparent and active demand of this EV for three
different days (represented by different line styles). It should
be noted that the EV demand is similar to a square waveform.
A total of 78 days were fully monitored (1-min average
samples, i.e., a total of 112,320). Fig. 4 shows the PDF of the
EV demand. When the EV battery is being charged, it is clear
that it typically demands 3.6kW. Although lower demand values exist, these occur when the battery is reaching full charge
(see in Fig. 3 the charging event represented by the dotted
line). This effect is ignored in this work. In terms of the power
factor, Fig. 5 highlights that the typical EV power factor is
0.98 (inductive, i.e., absorbing reactive from the grid).
B. Methodology to Create EV Profiles
Daily time-series EV profiles can be produced using the
PDFs presented in section II. The initial and final SOC define
how long the EV needs to be connected to the charging point.
Typically an EV draws 3.6 kW (0.98 inductive power factor)
of continuous demand. One segment (2 kWh) of charge in a

Accepted Paper

Accepted Paper
x 10

-3

4
Apparent
Active
Apparent
Active
Apparent
Active

3.5

EV Demand (kVA)

3
2.5
2

3.5

EV Demand (kVA)

1.5
1
0.5

3
2.5
2

1
0.5

0
0h

2h

4h

6h

8h

0
0h

10h 12h 14h 16h 18h 20h 22h 24h

Time of Day

Figure 3. Three examples of monitored active and apparent EV demand.

4h

6h

8h

10h 12h 14h 16h 18h 20h 22h 24h

Time of Day

1.4

Diversified EV Demand (kVA)

80

Probability (%)

2h

Figure 6. Example of individual EV profiles Weekday.

100

60
40
20
0

EV Load 1
EV Load 2
EV Load 3

1.5

1.5

2.5
3
EV Demand (kW)

3.5

1
0.8
0.6
0.4
0.2
0
0h

Two Connections
One Connection

1.2

2h

4h

6h

8h

10h 12h 14h 16h 18h 20h 22h 24h

Time of Day

Figure 7. Example of diversified EV demand Weekday.

Figure 4. PDF of the real EV demand monitored in the project.

1.4

Diversified EV Demand (kVA)

80
70

Probability (%)

60
50
40
30
20
10
0

1
0.8
0.6
0.4
0.2
0
0h

0.9

0.92

0.94
0.96
Power Factor

0.98

Two Connections
One Connection

1.2

2h

4h

6h

Figure 5. PDF of the real EV power factor monitored in the project.

Nissan LEAF needs approximately 40 minutes (3.6 kW).

With the above considerations, it is possible to create stochastic, realistic and detailed EV profiles to be used in EVrelated studies. The next steps are required for each profile:
1. Random selection of the number of connections using Table I. Then, for each connection follow steps 2 to 5.
2. Random selection of the start charging time using Fig. 1
and Fig. 2.
3. Random selection of the initial SOC using Table II.
4. Random selection of the final SOC using Table II (larger
than that of step 3).
5. Calculation of the time needed from the initial SOC to the
final SOC based on the required number of segments to be
charged (final SOC minus initial SOC).

8h

10h 12h 14h 16h 18h 20h 22h 24h

Time of Day

Figure 8. Example of diversified EV demand Weekend.

From the above steps, the charging process (i.e., a constant

3.6 kW demand with a 0.98 power factor) for each connection
occurs between the connection time (step 2) and finishing time
(step 2 + step 5). If there are two connections, each charging
process must not overlap with the other.
C. Example of EV Profiles
For weekdays, Fig. 6 shows examples of individual EV
profiles. As observed, these EV profiles represent the connection of EVs at different times of the day. Moreover, they have
different charging needs (duration). Finally, they consider one
car connecting twice in the same day (EV Load 2).
Fig. 7 and Fig. 8 further present the diversified demand for
1000 profiles for both weekday and weekend, respectively.
The diversified demand corresponds to the aggregated demand
divided by the number of profiles. As expected, the diversified
demand shows a similar behavior as that of the start time

Accepted Paper

1.4

1
0.8
Winter
Spring
Summer
Autumn
Overall

0.4
0.2
0

2h

4h

6h

8h

10h 12h 14h 16h 18h 20h 22h 24h

Time of day

Figure 9. Monitored diversified EV demand per season Weekday.

(Fig. 1 and Fig. 2). Fig. 7 and Fig. 8 also compare the diversified demand when considering one and two connections. For
weekdays, it is clear that the latter does not affect the evening
peak (~1.2 kW in both cases) but it does affect the weekday
morning peak (from 0.58 to 0.91 kW).
Crucially, it is clear that two connections have a significant
effect on the overall daily energy consumption. While the daily energy consumption considering a single connection during
weekdays and weekends is 13.9 and 13.7 kWh, this value increases to 17.8 and 16.9 kWh when two connections are considered for the same type of days. As previously highlighted,
this analysis is unique given that previous EV studies have not
explored multiple charging events.
D. PDFs For Different Seasons
No significant variance in the charging behavior of EVs
across seasons was found; the above analysis considered the
whole year. The analysis included both weekdays and weekends. To demonstrate the limited impact of seasonality, the
diversified EV demand for different seasons during weekdays
is shown here. Similar analysis for weekends is provided in
[10, 11] as part of the My Electric Avenue project.
Fig. 9 shows the average diversified EV demand for 100
sets of 1000 EV profiles for each season, as well as the yearly
(overall) charging behavior (PDF created in section II). Fig. 9
clearly shows no significant change in the EV demand across
seasons. If the overall diversified peak-demand (1.20 kW) is
considered as a reference, then the peak difference for each
season is: -0.68, 0.89, -1.80 and 3.69% for winter, spring,
summer and autumn, respectively. In terms of energy consumed, seasonal differences were found; 2.11, -2.67, -2.90 and
2.20%, considering again as a reference the diversified energy
consumed for the whole year (17.92 kWh).
IV.

ANALYSIS OF EV PROFILES AND IMPLICATIONS ON

UK DISTRIBUTION NETWORKS

A. Diversified Peak Demand and Coincidence Factor

The diversified peak demand (i.e., also known as the after
diversity maximum demand) is typically used in the design of
the electricity networks [12]. This section quantifies the diversified peak demand for different numbers of EVs (from 0 to
200 EVs) in a Monte Carlo approach. For each one, 100 random selections (from the pool of 1000 EVs) are carried out
and the corresponding diversified peak demand is calculated.
Fig. 10 and Fig. 11 highlights that the lower the number of

4
3.5
3
2.5
2
1.5
1
0.5
0

50

100
Number of EVs

150

200

Figure 10. Diversified peak EV demand for various EV numbers Weekday.

Diversified Peak EV Demand (kVA)

0.6

4
3.5
3
2.5
2
1.5
1
0.5
0

50

100
Number of EVs

150

200

Figure 11. Diversified peak EV demand for various EV numbers Weekend.

1

Coincidence Factor (p.u.)

Demand (kW)

1.2

Diversified Peak EV Demand (kVA)

Accepted Paper

Weekday
Weekend

0.9
0.8
0.7
0.6
0.5
0.4
0

50

100
Number of EVs

150

200

Figure 12. Coincidence Factor for various EV numbers.

EVs, the higher the diversified peak demand. Crucially, it can

be seen here that the diversified peak EV demand for more
than 50 EVs decreases slowly. As expected this approaches a
value of circa 1.2kW for large number of EVs.
Given the diversified peak demand of EVs, it is possible to
determine the coincidence factor among EVs, i.e., the diversified peak demand per individual EV. For both weekdays and
weekends, Fig. 12 highlights that the higher the number of
EVs, the lower the coincidence factor of EVs. Although not
shown in Fig. 12, the coincidence factor for a large number of
EVs (e.g., 1000) is 0.33 (i.e., 1.2 kW / 3.6 kW = 0.33).
B. Net Demand Analysis
The adoption of EVs is expected to increase the net demand in electricity networks. Fig. 13 and Fig. 14 show the
diversified demand of 1000 households (domestic unrestricted) in the UK for a typical weekday and weekend in winter -

Accepted Paper

Accepted Paper
2.5
2

Demand kVA)

VI.

Residential Load
EV Load
Total

This paper has presented results of a thorough statistical

analysis of the charging behavior of 221 real residential EV
users (Nissan LEAF, i.e., 24kWh, 3.6 kW demand and 0.98
inductive power factor) spread across the UK and monitored
over one year (68,000+ samples). PDFs of the number of connections per day (overlooked in most studies), start charging
time, initial SOC, and final SOC (per connection) for both
weekdays and weekends have been created.

1.5
1

0.5
0
0h

2h

4h

6h

8h

10h 12h 14h 16h 18h 20h 22h

Time of Day

24h

Figure 13. Diversified winter weekday profile: Residential + EV Demand.

2.5
Residential Load
EV Load
Total

Demand kVA)

2
1.5
1

0.5
0
0h

2h

4h

6h

8h

CONCLUSIONS

10h 12h 14h 16h 18h 20h 22h

Time of Day

24h

It has been shown that approximately 70% of the EVs are

connected once a day, irrespective of weekday and weekend.
In terms of the start charging time, EV users do change the
start charging time from weekdays to weekends and the start
charging time follows the UK residential load curve. The first
connection typically happens (for more than 70% of the EVs)
when the SOC is between 25 and 75%. Approximately 65% of
the EVs finish their first connection with a full battery. Second
connections normally occur with higher SOC, but disconnections before EVs are fully charged are more frequent.
A methodology that uses these unique PDFs to create stochastic, realistic and detailed EV profiles for impact and/or
Smart Grid-related studies has also been proposed. It has been
shown that multiple daily connections do not impact the evening peak, but this behavior does affect the morning peak as
well as the overall energy consumption, compared to most of
the studies that only consider single daily connections.

Figure 14. Diversified winter weekend profile: Residential + EV Demand.

REFERENCES

(e.g., January, maximum demand in the UK). These residential profiles were created using [13]. Fig. 13 and Fig. 14 also
present the weekday and weekend diversified demand of 1000
EVs and the net (residential + EV) demand. Irrespective of the
type of day, the diversified peak demand increases from 0.8
(without EVs) to about 2 kW when all houses have one EV.
This means an increase of more than 100%.

[1] Committee on Climate Change, "Reducing the UKs carbon footprint

and managing competitiveness risks," http://www.theccc.org.uk/wpcontent/uploads/2013/04/CF-C-Summary-Rep-web1.pdf, Apr. 2013.
[2] J. A. P. Lopes, F. J. Soares, and P. M. R. Almeida, "Integration of
electric vehicles in the electric power system," Proceedings of the
IEEE, vol. 99, no. 1, pp. 168-183, Jan. 2011.
[3] P. Richardson, M. Moran, A. Maitra, J. Taylor, and A. Keane, "Impact
of electric vehicle charging on residential distribution networks: An
Irish demonstration initiative," in CIRED, Stockholm, 2013.
[4] Z. Darabi and M. Ferdowsi, "Aggregated Impact of Plug-in Hybrid
Electric Vehicles on Electricity Demand Profile," IEEE Trans.
Sustainable Energy, vol. 2, no. 4, pp. 501-508, Oct. 2011.
[5] K. Petrou, J. Quirs-Torts, and L. F. Ochoa, "Controlling electric
vehicle charging points for congestion management of UK LV
networks," in IEEE PES ISGT, Washington, USA, 2015, pp. 1-5.
[6] J. Quirs-Torts, L. F. Ochoa, S. Alnaser, and T. Butler, "Control of EV
Charging Points for Thermal and Voltage Management of LV
Networks," IEEE Trans. on Pow. Syst., In Press (accepted Aug. 2015).
[7] J. Quirs-Torts, L. F. Ochoa, A. Navarro-Espinosa, M. Gillie, and R.
Hartshorn, "Probabilistic Impact Assessment of Electric Vehicle
Charging on Residential UK LV Networks," in CIRED, 2015, pp. 1-5.
[8] Nissan LEAF. Available: http://www.nissan.co.uk/
[9] My Electric Avenue. Available: http://myelectricavenue.info/
[10] J. Quirs-Torts and L. F. Ochoa, "Work Activity 3 Model Validation
and Data Analysis Report for Deliverables 3.1, 3.2, 3.3. and 3.4,"
The University of Manchester, "My Electric Avenue" Project,
http://myelectricavenue.info/, Dec. 2014.
[11] J. Quirs-Torts and L. F. Ochoa, "Work Activity 4 Business as Usual
Deterministic Impact Studies Report for Deliverables 4.1 and 4.2,"
The University of Manchester, "My Electric Avenue" Project,
http://myelectricavenue.info/, Jun. 2015.
[12] Scothish and Souther Energy, "Specification for planning & design of
greenfield LV housing estates," 2003.
[13] I. Richardson, M. Thomson, D. Infield, and C. Clifford, "Domestic
electricity use: A high-resolution energy demand model," Energy and
Buildings, vol. 42, no. 10, pp. 18781887, Oct. 2010.

V.

DISCUSSION

The PDFs presented above represent the charging behavior

of residential EV users. The behavior of a small set of commercial EV users has also been analyzed in the project (see
[11]). The residential and commercial charging behavior can
be potentially combined to create demand scenarios to be used
in planning of future UK electricity networks.
To create the PDFs presented in section II, this work has
investigated charging days only, thus resulting in the highest
EV demand. However, it is expected that some EV users will
not charge their vehicle every day (indeed, some days no EV
may be charged). This information (i.e., the percentage of EVs
charging during the same day) can be used to create different
scenarios to be used in different EV studies (e.g., provision of
reserves to the national or regional system operator).
This work has quantified the changes in the net demand for
large number of loads (1000 residential and 1000 EV loads).
However, care should be taken when studying low voltage
networks as the number of customers can be much lower.

Accepted Paper